It is time again for the year’s best sports event: the NCAA college basketball tournament. 65 teams in a single elimination tournament held over 3 successive weekends, with the winner taking all. Picking the teams for the tournament is a favorite conversation topic over coffee and beers (not that most of us can distinguish among the 336 eligible teams!). Joel Sokol, Paul Kvam and my research and business partner George Nemhauser have a system entitled the “Linear Regression/Markov Chain” (LRMC) to rank teams. You can get the pre-tournament rankings here. These rankings suggest that the four top teams in the land are North Carolina, Kansas, UCLA, and Texas A&M. The tournament committee picked Ohio State and Florida (number 5 and 6) instead of UCLA and Texas A&M.
If you are filling out your bracket, you can use the rankings to guide you (just pick the highest ranked team at each step). As an article in the Atlanta Journal Constitution writes:
LRMC’s most striking predictions this year: Picking No. 12 seed Arkansas over No. 5 seed Southern Cal and No. 10 seed Georgia Tech to knock off No. 7 seed UNLV.
The tournament committee this year had access to LRMC:
Nemhauser, a former Tech faculty athletics representative, arranged to provide a special version of LRMC to the selection committee for the first time this year. (The committee wouldn’t look at a ratings system that considered victory margin, so it got one with that component factored out.)
“It’s certainly one tool people could look at,” committee chairman Gary Walters said, but he went on to praise the RPI and the Sagarin ratings and to call the RPI “our primary quantitative tool.”
It will be interesting to see how some of the predicted upsets work out: perhaps the tournament committee will need a new primary quantitative tool.